paddlespeech.t2s.audio.spec_normalizer module

This modules contains normalizers for spectrogram magnitude. Normalizers are invertible transformations. They can be used to process magnitude of spectrogram before training and can also be used to recover from the generated spectrogram so as to be used with vocoders like griffin lim.

The base class describe the interface. transform is used to perform transformation and inverse is used to perform the inverse transformation.

check issues: https://github.com/mozilla/TTS/issues/377

class paddlespeech.t2s.audio.spec_normalizer.LogMagnitude(min=1e-05)[source]

Bases: NormalizerBase

This is a simple normalizer used in Waveglow, Waveflow, tacotron2...

Methods

inverse
transform

inverse(x)[source]

transform(x)[source]

class paddlespeech.t2s.audio.spec_normalizer.NormalizerBase[source]

Bases: object

Methods

inverse
transform

inverse(normalized)[source]

transform(spec)[source]

class paddlespeech.t2s.audio.spec_normalizer.UnitMagnitude(min=1e-05)[source]

Bases: NormalizerBase

This is the normalizer used in the

Methods

inverse
transform

inverse(x)[source]

transform(x)[source]